SGLang

SGLang is a model-serving engine with an OpenAI-compatible service surface and its own runtime-native graph/DSL features. LychD binds the declared service through the Soulstone adapter; it does not require the caller to depend on sgl.gen or any other SGLang-specific language.

Adapter-specific flags remain part of the exact engine profile. They must be declared, pinned, and included in the runtime receipt rather than inferred from a generic OpenAI-compatible endpoint.

The exact image, model, quantization, GPU and driver, launch arguments, readiness, inference, and shutdown still require an operator receipt. The current delivery boundary is recorded under SGLang integration.

See Soulstone Disciplines, SGLang, and the OpenAI-compatible connector boundary.